Most CFOs overspent on AI but expect payoff.
CFOs overshoot budgets but remain optimistic.
A new CFO Index from software firm Pigment reveals that most finance leaders have been surprised by the cost of recent AI deployments. The third-quarter index, based on 2,000 finance executives across the US, UK, France and Germany, found that 83% of respondents said their spending on consumption-based AI tools exceeded their budget plans ([1]). The average overrun was around 27%, with over one-third of companies overspending by more than 25% ([2]). Despite these budget busts, CFOs have not lost faith in AI’s potential. Survey participants rated their confidence in AI’s ability to deliver business value at an average of 8.2 out of 10 ([3]), indicating that finance chiefs largely expect payoffs down the line even if many have yet to see net gains in financial results.
These findings underscore the tension many organisations face as they ramp up AI initiatives. Rapid adoption of generative AI and other tools has introduced unpredictable, usage-based costs that are hard to forecast and control ([4]) ([5]). In the Pigment survey, 40% of finance leaders cited the unpredictability of usage as their top cost-management challenge ([6]). The typical budget overrun ranged from 11% to 25%, and in some cases more than 50% ([7]). Yet rather than slashing investment, most CFOs appear to be pressing forward, aiming to capture future benefits. The CFOs in the survey plan to increase their AI budgets for next year, albeit more modestly, by an average of roughly 15%, which is still less than the current year’s overrun, suggesting that many firms are reaching the limits of year-on-year AI spending growth ([8]). In practice, the optimism about AI’s promise is tempered by a recognition that costs must be kept in check for that promise to materialise in the bottom line.
CFO survey highlights AI cost governance gaps.
Finance leaders themselves are raising red flags about the need for tighter cost control and clearer ROI metrics for AI projects. Deloitte’s third-quarter 2026 CFO Signals survey of 200 North American CFOs underscores that while finance is taking a central role in AI decision-making, many organisations lack the tools for real-time cost management ([1]) ([2]). Nearly 80% of CFOs said their finance teams now have a formal role in approvals or evaluation of AI-related initiatives ([3]). Yet when asked how quickly they could rein in an AI project that was overspending, 74% responded that it would typically take several business days to weeks to pause the offending system ([4]). Only 7% have automated or near-instant shutoff mechanisms to curb runaway costs ([5]). This lag in response time exposes companies to substantial budgetary risk.
The survey points to other emerging safeguards. It found that two-thirds of CFOs now insist on measurement-focused funding processes for technology investments ([6]). About a quarter of finance chiefs primarily use stage-gate funding, releasing money in tranches only when projects meet defined milestones, and a similar share require a quantified business case and proven return on investment before approving major AI spend ([7]). However, roughly one in four CFOs say their companies still pursue technology projects due to top-down pressure without rigorous ROI vetting ([8]). The overall picture is one of a transition: after years of blank-cheque AI projects, boards and finance leaders are now shifting to an “ROI or stop” mindset. As Deloitte’s survey authors note, finance organisations will need more sophisticated, real-time cost governance to manage the rising complexity and spending volatility of AI at scale ([9]) ([10]).
A bank finds AI payoff in efficiency gains.
One area where the returns on AI are starting to become tangible is in operational efficiency. Valley National Bank, a regional US lender with $67 billion in assets, reported that its third-quarter efficiency ratio, a key measure of costs versus revenue, dropped to 49.4%, the bank’s lowest in five years ([1]). The bank’s CFO, Travis Lan, attributed this record efficiency performance to technology investments and the adoption of AI across both customer-facing and internal processes ([2]). A year ago, Valley’s efficiency ratio stood at 53.4%, so the improvement suggests significant cost savings and productivity gains in areas such as customer service, loan processing and back-office operations.
Valley National’s experience aligns with its leadership’s deliberate approach to AI deployment. Executives have emphasised focusing on measurable value and strong cost discipline in their AI strategy ([3]) ([4]). The bank was an early mover in applying AI to specific tasks, from automating customer support queries to speeding up fraud reviews and loan processing, with careful governance and cost controls in place ([5]) ([6]). By targeting well-defined processes and monitoring the impact, Valley has started to realise concrete financial benefits, such as improved productivity and lower operating expense ratios. This concrete example offers a note of encouragement to CFOs: with the right focus on accountability and measurement, AI’s return can emerge in the form of lower costs and efficiency gains, even if broad-based P&L impact remains a work in progress.